Weighted cohomology of arithmetic groups (Q1808855)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Weighted cohomology of arithmetic groups
scientific article

    Statements

    Weighted cohomology of arithmetic groups (English)
    0 references
    0 references
    14 May 2001
    0 references
    Let \(\mathbb{G}\) be a semisimple algebraic group defined over the rational numbers, \(K\) a maximal compact subgroup of \(G=\mathbb{G}(\mathbb{R})\), and \(\Gamma\subset \mathbb{G}(\mathbb{Q})\) a neat arithmetic subgroup. Let \(X =\Gamma\setminus G/K\) be the locally symmetric space associated to \(\Gamma\), and \(\mathbb{E}\) the local system on \(X\) constructed out of a finite-dimensional, irreducible, algebraic representation \(E\) of \(\mathbb{G}\). Weighted cohomology is an invariant of \(\Gamma\) introduced by \textit{M. Goresky, G. Harder}, and \textit{R. MacPherson} [Invent. Math. 116, 139-213 (1994; Zbl 0849.11047)] in the study of the trace of Hecke operators in the cohomology of \(\Gamma\). For each \(p\in X^*(A)\otimes\mathbb{Q}\) there are weight profiles \(\overline{p}\) and \(\underline{p}\) and groups \(W^{\overline{p}}H^i(\Gamma,E)\) and \(W^{\underline{p}} H^i(\Gamma,E)\). Let the space \(X\) be a Riemannian manifold, a complete metric coming from an invariant metric on \(G/K\). There is a natural choice of inner product on \(E\) metrizing \(\mathbb{E}\). For each \(\lambda\in{\mathfrak A}^*= X^*(A)\otimes\mathbb{R}\), \textit{J. Franke} [Ann. Sci. Éc. Norm. Supér. (4) 31, 181-279 (1998; Zbl 0938.11026)] has defined weighted \(L^2\) cohomology groups via complexes of forms on \(X\) satisfying certain square-integrability conditions given by \(\lambda\). For very positive \(\lambda\) these give the full cohomology \(H^i(X,\mathbb{E})\) and for very negative \(\lambda\) they give \(H^i_c(X,\mathbb{E})\). The definition of these groups involves a little reduction theory. The main theorem is that these two cohomology theories are the same: Theorem A. Let \(\lambda = -p -\rho_0\). There are natural isomorphisms \[ \begin{aligned} & W^{\overline{p}}H^i(\Gamma,E)\simeq H^i_{\lambda-\log}(X,\mathbb{E}),\\ & W^{\underline{p}}H^i(\Gamma,E)\simeq H^i_{\lambda+\log}(X,\mathbb{E}).\end{aligned} \] The two assertions are equivalent by duality. Consider the space \(S_{\lambda-\log}(\Gamma\setminus G)\) (resp. \(S_{\lambda+\log}(\Gamma\setminus G))\) of smooth complex-valued functions \(f\) on \(\Gamma\setminus G\) such that, for all \(j\geq 0\) (resp. some \(j\leq 0\)), \[ w_\lambda\log (w_\varepsilon)^j R_Df\text{ is square-integrable} \] for every operator \(D\) in the universal enveloping algebra of \({\mathfrak G} =\text{Lie}(G)\). They are \(({\mathfrak G},K)\)-modules. A theorem of Franke computes \(H^i_{\lambda\pm\log}(X,\mathbb{E})\) in terms of \(S_{\lambda\pm\log}(\Gamma\setminus G)\), and so Theorem A is equivalent to: Corollary A. Let \(\lambda = -p -\rho_0\). There are natural isomorphisms \[ \begin{aligned} & W^{\overline{p}}H^i(\Gamma,E)\simeq H^i_{{\mathfrak G},K} (S_{\lambda-\log}(\Gamma\setminus G)\otimes E),\\ & W^{\underline{p}}H^i(\Gamma,E)\simeq H^i_{{\mathfrak G},K} (S_{\lambda+\log}(\Gamma\setminus G)\otimes E).\end{aligned} \] There are two special profiles, upper middle \(\mu = \overline{-\rho_0}\) and lower middle \(\nu= \underline{-\rho_0}\). They are dual to each other. Theorem A says that \(W^\mu H^i(\Gamma,E)\simeq H^i_{(2)-\log}(X,\mathbb{E})\) and \(W^\nu H^i(\Gamma,E)\simeq H^i_{(2)+\log}(X, \mathbb{E}).\) The relation of \(H^i_{(2)\pm\log}(X,\mathbb{E})\) to the usual \(L^2\) cohomology is given by: Theorem B. If the ranks of \(G\) and \(K\) are equal then \[ H^i_{(2)-\log}(X,\mathbb{E})\simeq H^i_{(2)}(X, \mathbb{E})\simeq H^i_{(2)+\log}(X,\mathbb{E}). \] In fact this holds under a slightly more general condition from \textit{A. Borel} and \textit{W. Casselman} [Duke Math. J. 50, 625-647 (1983; Zbl 0528.22012)]. Theorems A and B imply: Corollary B. If the ranks of \(G\) and \(K\) are equal then \[ W^\mu H^i(\Gamma,E)\simeq H^i_{(2)}(X,\mathbb{E})\simeq W^\nu H^i(\Gamma,E). \] This corollary is related to Zucker's conjecture (a theorem thanks to Looijenga and Saper-Stern). Suppose that \(X\) is Hermitian and \(\widehat{X}\) is its Baily-Borel Satake compactification; \(\widehat{X}\) is a complex projective variety. A theorem of Goresky-Harder-MacPherson says that \(W^\mu H^i(\Gamma,E)\) and \(W^\nu H^i(\Gamma,E)\) are isomorphic to the intersection cohomology \(IH^i(\widehat{X},\mathbb{E})\). The corollary then implies Zucker's conjecture. The proof of Theorem A has three steps. The first step is the equivalence with Corollary A (due to Franke) and is dealt with in Section 1. The second step is to show that there is an inclusion \({\mathbf W}^{\overline{p}}{\mathbf C}^\bullet(\mathbb{E})\hookrightarrow{\mathbf S}^\bullet_{\lambda-\log}(\mathbb{E})\) for \(\lambda =-p -\rho_0\). The final step is to show that this inclusion is a quasi-isomorphism via a local calculation on \(\overline{X}\). There are two key points here: Proposition 3.2 (reduction to the space of \(N_P\)-invariants of \({\mathbf S}^\bullet_{\lambda-\log}(\mathbb{E})\) over neighbourhoods of boundary points, (a version of) an idea that goes back to van Est) and Lemma 3.5 (where the logarithmic terms in (*) become essential). In Section 0, for the reader's convenience, the proof of Theorem A is sketched in the simplest possible case, namely when \(\mathbb{G}= \text{SL}(2)\) and \(E\) is trivial. The proof of Theorem B, which is global and independent of the previous sections, is in Section 4 and uses results of Franke and Borel-Casselman [loc. cit.] and standard Lie algebra cohomology arguments. The final section (\S 5) contains applications and several comments on relations to other work and on generalizations.
    0 references
    0 references
    0 references
    0 references
    0 references